cyberivy
Google ArtemisMinitapOpen Source AIApache 2.0Android AgentsAI AgentsSoftware ProvenanceDeveloper Tools

Minitap accuses Google of omitting open-source credit in Artemis

September 12, 2026

Vergleichsgrafik aus dem Artemis-Repository mit Balken fΓΌr mehrere Android-Agenten und deren AndroidWorld-Ergebnisse

Minitap documents identical code, prompts, and a removed author list in Google's Artemis. The dispute shows why provenance records are essential for open AI tools.

What this is about

Paris-based startup Minitap accuses Google of incorporating substantial parts of its Apache 2.0-licensed mobile-use project into Artemis, a tool for automating Android devices, without visibly crediting the original developers. Minitap published its detailed account on September 12, 2026. The allegation goes beyond similar ideas: the company points to word-for-word code, identical agent instructions, the same unusual example, and an author list that named three Minitap engineers in an older Artemis version before those names were replaced.

Google had not published a detailed public response when we checked. A change restoring credit was, however, opened in the Artemis repository. This makes part of the core record publicly inspectable while the intent and internal decision-making remain unknown.

What Artemis and mobile-use actually do

Both projects let AI agents operate Android devices. They connect to a phone or emulator, inspect the visible state, and perform actions such as tapping, swiping, or opening an app. Minitap identifies several concrete overlaps: an ADB tunnel, instructions for an agent called Hopper, and a WhatsApp example that sends New Year messages to Alice, Bob, and Charlie.

The version history is especially relevant. An older Artemis pyproject.toml file listed Pierre-Louis Favreau, Jean-Pierre Lo, and Nicolas Dehandschoewercker as authors. A later version replaced those names with another author. GitHub also records a force push in August. Minitap additionally points to a results-file bug that it says was present in both implementations for a time. These indicators support a technical provenance link, but they do not yet explain who made each decision or why.

Why it matters

Apache 2.0 explicitly permits use, modification, and commercial redistribution. Freedom does not mean that provenance can simply disappear. Section 4 of the license requires, among other things, preservation of applicable copyright and attribution notices and identification of modified files. The precise obligations here depend on which files and notices were carried over; there is no final legal assessment.

For open AI projects, the practical consequence is larger than a dispute over names. Provenance tells developers where to report bugs, who understands design choices, and which components were independently created. If small teams expect their work to reappear under a large company's name, they have less reason to share early research in public. At the same time, it would be misleading to describe Artemis as entirely copied: Minitap itself says the Google project contains original engineering work too.

In plain language

Imagine a neighbor publishes a bread recipe and explicitly allows anyone to modify it or sell the resulting bread. A large bakery may build on that recipe. It should not remove the note showing where the recipe came from and imply that every step originated in its own kitchen. This dispute is about that provenance note.

A practical example

A small team releases 5,000 lines of code for a mobile testing agent. A larger company imports 800 lines, changes 200 of them, and adds 10,000 original lines. The new product is mostly independent by volume. The imported 800 lines and their license notices still matter. A clear README entry, preserved copyright notices, and an auditable change history would tell future developers which bugs came from which source. Without that trail, the smaller team must recover old commits, match unusual examples, and identify shared bugs simply to establish what it created.

Scope and limits

First, the detailed case was assembled by Minitap, an involved party. The linked GitHub revisions make verification possible but do not replace an independent investigation. Second, it is not known why the author names were removed or whether this reflected a deliberate decision by the responsible Google team. Third, matching components do not prove that Artemis as a whole was copied; Minitap explicitly acknowledges Google's own engineering contributions.

This article is also not legal advice. Whether the facts amount to a license violation would require expert review of all files, notices, and distribution paths. What is established for now is narrower: publicly documented provenance was unclear, and Google responded to the complaint with a credit-related change.

SEO & GEO keywords

Google Artemis, Minitap mobile-use, open-source attribution, Apache 2.0, Android agents, AI agents, GitHub history, software provenance, author credit, open-source license

πŸ’‘ In plain English

Minitap shows publicly inspectable overlaps between its open Android agent and Google's Artemis. It remains unclear why developer names were removed and whether license obligations were breached.

Key Takeaways

  • β†’Minitap documents word-for-word code, identical prompts, and the same unusual example.
  • β†’An older Artemis file named three Minitap engineers, while a later version did not.
  • β†’Google opened a change restoring credit after the public complaint.
  • β†’Intent, responsibility, and any possible license breach remain unresolved.
  • β†’Clear provenance protects trust and maintainability in the open-source ecosystem.

FAQ

What is Google Artemis?

Artemis is an open Google project for letting AI agents perform and test tasks on Android devices.

What does Minitap accuse Google of?

Minitap cites identical implementations, prompts, examples, and removed developer names. It is asking for clear provenance credit.

Has a license breach been proven?

No. The technical overlaps are documented, but a final legal review and a complete Google response are still missing.

Why does attribution matter in practice?

It preserves traceable ownership, maintenance paths, and the incentive to publish research openly.

Sources & Context